The rapid growth of artificial intelligence is putting real pressure on data center infrastructure, especially when it comes to moving huge amounts of data between processors. Here, we’ll take a look at a new collaboration between Lightmatter and Cadence that tries to tackle these challenges using co-packaged optics (CPO).
They’re combining advanced silicon photonics with top-tier electronic design automation (EDA) and interconnect IP. The goal? To enable the next generation of AI-scale systems.
Why Co-Packaged Optics Matter for AI Infrastructure
AI models are getting bigger and more complex every year. Traditional electrical interconnects just can’t keep up with the bandwidth and energy demands anymore.
Pluggable optics and near-package solutions have squeezed out more performance for a while, but they’re hitting their limits.
Co-packaged optics bring optical and electrical components together in one package, making interconnects much shorter. This setup cuts power use, boosts signal integrity, and allows a lot more bandwidth in a tight space.
For hyperscale AI data centers, these capabilities are quickly becoming non-negotiable.
The Lightmatter–Cadence Collaboration
On January 26, 2026, Lightmatter and Cadence announced they’d team up on developing manufacturing-ready CPO solutions. Both companies bring something valuable—Lightmatter’s photonics know-how, Cadence’s strengths in packaging and electronic design.
The main technical goal is to combine Cadence’s high-speed SerDes IP with Lightmatter’s Passage optical engine. Together, they want to create fully integrated, 3D-stacked photonic interconnects built for AI workloads.
Lightmatter’s Passage Optical Engine
Lightmatter calls Passage the world’s first 3D-stacked silicon photonics engine. It’s designed to connect thousands—or even millions—of XPUs, which tackles one of AI’s toughest problems: moving data efficiently.
Stacking photonic and electronic parts in three dimensions lets Passage reach new levels of interconnect density, all while keeping power use in check.
This kind of architecture could really change how we scale AI systems past today’s limits.
Cadence’s Role: IP, EDA, and UCIe
Cadence brings decades of experience in electronic design automation and advanced-node CMOS integration. A big part of the partnership is Cadence’s Universal Chiplet Interconnect Express (UCIe) IP, which supports breaking systems down into smaller chiplets.
UCIe lets designers mix and match chiplets across different process nodes and functions. That fits naturally with co-packaged optics and the trend toward heterogeneous integration.
Moving Beyond Pluggable and Near-Package Optics
Leaders at both companies are clear—this isn’t just about small improvements. They’re aiming to leap past pluggable and near-package optics and deliver fully integrated optical-electrical solutions ready for mass production.
That kind of integration is key for the bandwidth and energy efficiency future AI models will need, especially as hyperscale architectures get even more ambitious.
Industry Perspective: An Inevitable Shift
Some industry analysts say 3D-stacked photonic designs are basically inevitable at this point. Roy Chua from AvidThink went so far as to call the approach inevitable, arguing that the Lightmatter–Cadence partnership is laying the groundwork for seamless electrical-optical connectivity.
In his view, this isn’t just about one product—it’s a sign of where AI infrastructure design is going.
Implications for Hyperscale Data Centers
Hyperscale AI data centers are the main focus for this CPO platform. In these environments, performance-per-watt and scalability can make or break operational success.
Cadence executive Boyd Phelps points out that next-gen optical interconnects are crucial for getting the most out of performance and efficiency at scale.
Some of the key benefits expected from the collaboration include:
Looking Ahead
Lightmatter and Cadence are teaming up. They’re bringing together advanced CMOS IP, chiplet standards, and some pretty wild photonic packaging.
The goal? A production-ready CPO platform. For the AI world, that could be a real game-changer as models keep ballooning in size and capability.
Honestly, in the long run, I think fully integrated optical-electrical systems will form the backbone of AI infrastructure.
This partnership puts both companies right at the edge of that shift.
Here is the source article for this story: Lightmatter & Cadence: Advancing Co-Packaged Optics For Next-Gen AI Infrastructure